A year after fashion and beauty companies took to Instagram en mass to show support for the Black community and the Black Lives Matter movement, our analysis of 27,000 images posted by 34 brands showed that while many did increase the diversity of skin tones in their Instagram images, the increases were often only marginal. Light skinned models still prevail.
We made this readily apparent with interactive and static data visualization.
The piece was one of the more widely read items on our site and was especially well read by members of the fashion and beauty industry. Researchers of inequity and company representatives reached out asking us to share our data and methods so that they could bring better accountability to their organizations and study it further. Influencers shared the story and graphics with their followers. Fashion influencer Bryanboy called it “very essential reading” Later in the year our data and graphics were included in an episode of an episode of the The BoF Show on Bloomberg TV.
First we used custom built tools to collect and store Instagram posts using python and node. Then we constructed a database front-end that allowed us to evaluate and categorize every image we collected. That piece of software was written in node.
The visualizations have three modes to allow readers to explore the data. A timeline view, a clustered gradient view, and a combination of the two—a view of two clusters, split by whether the post was from before or after Blackout Tuesday. These three modes deftly showed how long brands stopped posting to Instagram during the US unrest, the distribution of skin tone depicted on a brand’s account, and how that distribution changed after Blackout Tuesday. In all three views, dots can be tapped or moused-over to reveal the image it represents.
We size-optimized the photographs using the command line tool imagemagick.
What was the hardest part of this project?
But that was just the start. We then used the software we wrote to evaluate each image by hand, establishing the number of people in the image, their skin colors, and whether or not the image was suitable for inclusion in our analysis.
What can others learn from this project?
Firstly, our project is a great example of how to hold organizations accountable through data. Second, it shows the opportunity for journalists to create data where none previously existed. There was no dataset of the skin tones of models promoted by fashion and beauty brands, despite the information being in plain view. We were willing to put in the work, and made a first-of-its kind dataset.